Getting Started

The beginning of 2020 when they have planned their own future direction is to ARM devices edge computing direction development, is expected in the next 3-5 years ARM platform will rise in various computing fields, from smart phones to small Arduino edge devices, to large supercomputing systems, in every corner of the world we can see the figure of ARM.

In order to promote SARS-Cov-2 vaccine research and development, they also formed a set of ARM64 general-purpose computing platform, the hardware used is Raspberry Pi 4, the system uses Ubuntu 20.04 aarch64, because the main focus of low power consumption, so the entire computing platform is running at home, 50 units of Raspberry Pi Power consumption is basically at 300-400 watts, at the same time, they are also setting up Kubernets under the machine learning platform, able to complete the training of models on ARM, of course, the performance and X86 + GPU can not be compared, overall they can learn to large distributed project training and cluster management experience on a large scale.

Then came the news that Nvidia was ready to acquire ARM, and once the news came out, I was even more determined that ARM was definitely the future of low-power general-purpose computing platforms in the future, although the ecology of ARM is not very perfect now, and many developers’ tool chains will have great difficulty in porting from X86 to ARM platforms, however, by the time Apple released its new M1 chip, all If ARM leads us to another fantasy CPU world, Apple is the leader of this tribe who leads us to break out of the enemy group, out of a road of their own, and this road will turn from a winding path to a highway to two worlds. Two days after the release of the new M1 chip, I carefully watched the official release video and immediately placed an order for a Mac Mini M1 and decided to use this productivity device to increase my income, edit videos, create apps and of course play with Tensorflow + Swift, although the machine will not be available until December. However, reading Twitter today made me feel like I made the right choice.

SoC architecture

Apple packages all the processing components, memory, processor, storage all in an M1 chip using a 5nm process.

From the figure, you can observe that both CPU and GPU share unified memory and embed the Neural Engine neural network processing engine. From the official introduction, we can learn that when processing images and videos, the processing speed is increased by 3-4 times compared to the previous generation. This alone makes me think that a Mac with M1 is worth buying, and also Tensorflow will definitely support M1 in the future, so when the time comes, we can test how many times the training time will be shortened when training deep learning models directly on the Mac Mini, compared to the previous The first generation of Mac Mini.

Apple brags that its M1 is the most advanced and lowest power consumption SoC, so let’s wait and see.

We’ll see what happens

golang on M1

Reference article link:

Running golang native on Apple M1

Build go-darwin-arm64-bootstrap on Raspberry Pi 4 arm64

git clone https://go.googlesource.com/go
git fetch https://go.googlesource.com/go refs/changes/58/272258/2 && git checkout -b change-272258 FETCH_HEAD
cd go/src
GOOS=darwin GOARCH=arm64 . /bootstrap.bash